tcp in wireless networks: challenges, optimizations and ...24680/fulltext02.pdf · during the last...

Division for Information TechnologyDepartment of Computer Science

Karlstad University Studies2005:13

Stefan Alfredsson

TCP in Wireless Networks: Challenges, Optimizations

and Evaluations

TCP in Wireless Networks: Challenges,Optimizations and Evaluations

This thesis presents research on transport layer behavior in wireless networks. As the Internet is expanding its reach to include mobile devices, it has become apparent that some of the original design assumptions for the dominant transport protocol, TCP, are approaching their limits. A key feature of TCP is the congestion control algorithm, constructed with the assumption that packet loss is normally very low, and that packet loss therefore is a sign of network congestion. This holds true for wired networks, but for mobile wireless networks non-congestion related packet loss may appear. The vary-ing signal power inherent with mobility and handover between base-stations are two example causes of such packet loss. This thesis provides an overview of the challenges for TCP in wireless networks together with a compilation of a number of suggested TCP optimizations for these environments. A TCP modification called TCP-L is proposed. It allows an application to increase its performance, in environments where residual bit errors normally give a degraded throughput, by making a reliability tradeoff. The performance of TCP-L is experimentally evaluated with an implementation in the Linux kernel. The transport layer performance in a 4G scenario is also experimentally investigated, focusing on the impact of the link layer design and its parameterization. Further, for emulation-based protocol evaluations, controlled packet loss and bit error generation is shown to be an important aspect.

Karlstad University StudiesISSN 1403-8099

ISBN 91-85335-53-3

Stefan Alfredsson T

CP in W

ireless Netw

orks: Challenges, O

ptimizations and E

valuations

Karlstad University Studies

2005:13

Stefan Alfredsson


and Evaluations

Stefan Alfredsson. TCP in Wireless Networks: Challenges, Optimizations and Evaluations.

Licentiate thesis

Karlstad University Studies 2005:13ISSN 1403-8099ISBN 91-85335-53-3

© The author

Distribution:Karlstad UniversityDivision for Information TechnologyDepartment of Computer ScienceSE-651 88 KARLSTAD SWEDEN+46 54-700 10 00

www.kau.se

Printed at Universitetstryckeriet, Karlstad 2005

TCP in Wireless Networks: Challenges,Optimizations and EvaluationsSTEFAN ALFREDSSONDepartment of Computer Science, Karlstad University

Abstract

This thesis presents research on transport layer behavior in wireless networks.As the Internet is expanding its reach to include mobile devices, it has becomeapparent that some of the original design assumptions for the dominant trans-port protocol, TCP, are approaching their limits. A key feature of TCP is thecongestion control algorithm, constructed with the assumption that packet lossis normally very low, and that packet loss therefore is a sign of network con-gestion. This holds true for wired networks, but for mobile wireless networksnon-congestion related packet loss may appear. The varying signal power inher-ent with mobility and handover between base-stations are two example causesof such packet loss. This thesis provides an overview of the challenges for TCPin wireless networks together with a compilation of a number of suggested TCPoptimizations for these environments. A TCP modification called TCP-L is pro-posed. It allows an application to increase its performance, in environmentswhere residual bit errors normally give a degraded throughput, by making a re-liability tradeoff. The performance of TCP-L is experimentally evaluated withan implementation in the Linux kernel. The transport layer performance in a4G scenario is also experimentally investigated, focusing on the impact of thelink layer design and its parameterization. Further, for emulation-based proto-col evaluations, controlled packet loss and bit error generation is shown to be animportant aspect.

Keywords: Wireless networks, TCP/IP, TCP-L, Network emulation

i

Acknowledgments

I would first and foremost like to thank my supervisor Anna Brunstrom. Her sup-port and encouragement during my research studies and article writings has beeninvaluable. I also thank my other co-authors, namely Annika Wennstrom, MikaelSternad and Johan Garcia (in order of appearance). My co-supervisor Tony Ot-toson deserves a special thanks, as does the other colleagues in the “Wireless IP”project and the Distributed Systems and Communications Research Group, forsupport and interesting discussions. Anna Widenius put a lot of effort into theconstruction and development of WIPEMU, which is greatly appreciated. Mycolleagues at the Computer Science department are well worth an acknowledge-ment. Thank you for providing such a nice and friendly working environment.Financial support has been provided (in various constellations) by the SSF, Vin-nova, CMIT, and the PCC++ research school. I acknowledge this with gratitude.

Last but not least, a big “thank you” to my family and friends for their love,support and patience, and for enduring my sometimes strange working hours.

Karlstad, May 2005

Stefan Alfredsson

iii

List of Appended Papers

This thesis is based on the work presented in the following five papers. Refer-ences to the papers will be made using the Roman numbers associated with thepapers.

Paper I Annika Wennstrom, Stefan Alfredsson, and Anna Brunstrom. TCPover Wireless Networks. Karlstad University Studies 2004:21, Karl-stad University, Sweden, May 2004.

Paper II Stefan Alfredsson and Anna Brunstrom. Bit Error Transparent Mul-timedia Transport. Chapter 10 in Perspectives on Multimedia - Com-munication, Media and Information Technology, Wiley 2004.

Paper III Stefan Alfredsson and Anna Brunstrom. TCP-L: Allowing Bit Errorsin Wireless TCP. In Proceedings of the IST Mobile & Wireless Com-munications Summit, Aveiro, Portugal, June 2003.

Paper IV Stefan Alfredsson, Anna Brunstrom, and Mikael Sternad. Emulationand Validation of a 4G System Proposal. To appear in Proceedings ofRadioVetenskap och Kommunikation, Linkoping, Sweden, June 2005.

Paper V Johan Garcia, Stefan Alfredsson, and Anna Brunstrom. The Impactof Loss Generation on Emulation-based Protocol Evaluation. Undersubmission.

Some of the papers have been subjected to minor editorial changes.

v

Contents

Abstract i

Acknowledgments iii

List of Appended Papers v

Introductory Summary 11 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Data Transmission Background . . . . . . . . . . . . . . . . . . 53 Overview of Research Area . . . . . . . . . . . . . . . . . . . . 84 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . 95 Research Methodology . . . . . . . . . . . . . . . . . . . . . . 106 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . 107 Paper Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 11

7.1 Paper I - TCP over Wireless Networks . . . . . . . . . . 117.2 Paper II - Bit Error Transparent Multimedia Transport . 117.3 Paper III - TCP-L: Allowing Bit Errors in Wireless TCP 127.4 Paper IV - Emulation and Validation of a 4G System

Proposal . . . . . . . . . . . . . . . . . . . . . . . . . 127.5 Paper V - The Impact of Loss Generation on Emulation-

based Protocol Evaluation . . . . . . . . . . . . . . . . 128 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Paper I: TCP over Wireless Networks 171 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192 Wireless Networks . . . . . . . . . . . . . . . . . . . . . . . . 21

2.1 Wireless LANs . . . . . . . . . . . . . . . . . . . . . . 212.2 Wireless WANs . . . . . . . . . . . . . . . . . . . . . . 22

vii

2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 293 TCP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.1 Slow Start and Congestion Avoidance . . . . . . . . . . 303.2 Fast Retransmit and Fast Recovery . . . . . . . . . . . . 323.3 TCP Options . . . . . . . . . . . . . . . . . . . . . . . 323.4 Other Mechanisms . . . . . . . . . . . . . . . . . . . . 333.5 TCP Variants . . . . . . . . . . . . . . . . . . . . . . . 34

4 Problems with TCP in Wireless Networks . . . . . . . . . . . . 355 Proposed Optimizations . . . . . . . . . . . . . . . . . . . . . . 36

5.1 Link Layer . . . . . . . . . . . . . . . . . . . . . . . . 365.2 Split Connection . . . . . . . . . . . . . . . . . . . . . 385.3 Explicit Notification . . . . . . . . . . . . . . . . . . . 395.4 End-to-end . . . . . . . . . . . . . . . . . . . . . . . . 415.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . 44

6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 45

Paper II: Bit Error Transparent Multimedia Transport 511 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532 Transport Protocol Background . . . . . . . . . . . . . . . . . . 54

2.1 TCP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552.2 TCP over Wireless Links . . . . . . . . . . . . . . . . . 562.3 TCP-L . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . 584 Experiment Results . . . . . . . . . . . . . . . . . . . . . . . . 59

4.1 Randomly Distributed Errors . . . . . . . . . . . . . . . 594.2 Impact of Burstiness . . . . . . . . . . . . . . . . . . . 604.3 Experiment Conclusions . . . . . . . . . . . . . . . . . 61

5 Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . 626 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 64

Paper III: TCP-L: Allowing Bit Errors in Wireless TCP 691 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 712 Header Decoding/Recovery . . . . . . . . . . . . . . . . . . . . 73

2.1 TCP Header Overview . . . . . . . . . . . . . . . . . . 742.2 Phases of Recovery . . . . . . . . . . . . . . . . . . . . 762.3 Recovery Algorithm . . . . . . . . . . . . . . . . . . . 78

3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.1 Experiment Setup . . . . . . . . . . . . . . . . . . . . . 793.2 Experiment Results . . . . . . . . . . . . . . . . . . . . 80

viii

4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

Paper IV: Emulation and Validation of a 4G System Proposal 871 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 892 WIPEMU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903 Transport Protocols . . . . . . . . . . . . . . . . . . . . . . . . 924 Experiment Overview . . . . . . . . . . . . . . . . . . . . . . . 945 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 976 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 102

Paper V: The Impact of Loss Generation on Emulation-based ProtocolEvaluation 1051 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1072 Network Emulation Background . . . . . . . . . . . . . . . . . 1093 Controlled Packet Loss . . . . . . . . . . . . . . . . . . . . . . 110

3.1 Positional Dependencies . . . . . . . . . . . . . . . . . 1113.2 Repeatability . . . . . . . . . . . . . . . . . . . . . . . 114

4 Controlled Bit-errors . . . . . . . . . . . . . . . . . . . . . . . 1174.1 Time-driven Bit-errors . . . . . . . . . . . . . . . . . . 1184.2 Data-driven Bit-errors . . . . . . . . . . . . . . . . . . 120

5 Implementation Details . . . . . . . . . . . . . . . . . . . . . . 1226 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

ix

Introductory Summary

1. Introduction 3

1 Introduction

During the last decades, computing has become more and more pervasive. Per-sonal computers are commonly found in homes, cars are controlled by micro-chips, and mobile phones have more computing power than an average PC adecade ago.

Closely related to computing is communication, for what use is a computa-tion if its result cannot be communicated? Either to the end user, via screen andkeyboard, or to another computing device. By enabling communication betweendevices, their power is combined, and their efficacy is multiplied many timesmore.

Communication in its simplest form consists of transmitting informationfrom one place to another, via a medium. In the case of computers, communica-tion is often performed via metallic mediums, such as copper on a circuit board,coaxial cable, or twisted pair wire, popular in the pervasive Ethernet technology.In the case of metallic medium, electrons are used to build up the signals. Glassfiber is also commonly used for communication, but here photons (light) areused instead of electrons. All these mediums require some form of mechanicalconnection between the communication end-points.

Starting with Marconi and Tesla some 100 years back, we also have theability to communicate without using wires, as electro-magnetic waves do notrequire any medium to be present. This allows for a great amount of commu-nication freedom, as phones and computers can communicate without having amechanical connection.

During the last decade, wireless communication equipment has become read-ily available for use with personal computers. Wireless local area networking iscommonly done with the IEEE 802.11 standard, or enhancements thereof [28].Wider area coverage is made possible by utilizing the existing mobile phone in-frastructure for data transmission in the form of for example GPRS [3, 11]. Theintegration between packet data and voice communications is further evolvedwith the deployment of 3G/UMTS [27]. This system is specifically designed tocarry packet data, video and voice communications with much higher capacitythan previous wide area coverage networks.

Parallel to this development is the expansion of the Internet, which has seenan explosive growth since the introduction of the World Wide Web in the early1990s. It seems like a natural evolution that the Internet should be accessiblewirelessly, just like the mobile phone is superseding the fixed phone. The num-ber of people with only a mobile phone subscription is increasing, and it can

4 Introductory Summary

be expected that this trend will continue as the call rates become cheaper andcheaper. Likewise more and more Internet users can be expected to use a wire-less instead of a wired connection.

The core of the Internet is the best effort IP packet network, which onlyforwards packets from one destination to another. On top of this are protocolsthat provide a data transport service, where the major reliable transport protocolused today is TCP [20]. Although TCP was designed over two decades ago, ithas survived the evolution in networking. For example, the bandwidth capacityhas increased many orders of magnitudes, while the size of the network has alsogrown immensely. Thanks to the extensibility of the protocol (see for example[12]) and the key idea that intelligence should be pushed to the network edge (the“end-to-end argument” [22]), TCP still performs very well with wired networktechnologies. Even as the Internet is used over wireless networks, it works, butthe limitations in its design is starting to show [6].

The pervasiveness of the Internet in combination with the increased use ofwireless technologies makes TCP over wireless an important research topic, andit is thus a background motivation for this thesis. The reported limitations in-spired a further analysis of the workings of TCP in wireless networks. Whattypes of networks are used, how does TCP behave in these, what optimizationshave been proposed? The result of this analysis is reported in Paper I. Earlierresearch on a partially reliable transport protocol [4], the UDP-Lite [13] proto-col, and the characteristics of wireless networks, inspired the investigation ofthe concept of allowing bit errors in TCP. Papers II and III report on the find-ings of this research. To evaluate the protocols, experimental network test-bedswere prepared. Within the test-beds, investigations were also performed on theinteractions between the lower layers and the transport layer. Paper IV presentsresearch on the impact of the link layer design and associated parameter settingson the transport layer. Another related aspect is considered in Paper V wherethe impact of controlled loss for network emulation on protocol evaluation isstudied.

The remainder of this introductory summary is organized as follows. Section2 contains a general background on data transmission, focused on the physicallayer. This is followed by an overview of the research area in section 3. Section4 outlines the research questions for the thesis. The research methodology isdiscussed in Section 5. This is followed by the main contributions of the thesisin Section 6. Section 7 provides a summary of the included papers. Finally,thoughts on future work are presented in Section 8.

2. Data Transmission Background 5

2 Data Transmission Background

Data transmission on the transport layer is extensively discussed in the includedpapers. Data transmission at the lower layers are not, however. Therefore, thissection provides a short introduction to modulation, the basic technology for datatransmission. This gives a better understanding for the workings of the lower lay-ers, and explains differences between wired and wireless communication. Thissection is aimed at readers not already familiar with these concepts.

To facilitate data transmission in a medium on the physical layer, a carriersignal is often used (Figure 1). This carrier is modulated, or modified, in certainways according to the data being transmitted. The most frequent modulationtechniques are amplitude, phase and frequency modulation. For example, com-mon broadcast radio transmissions used amplitude modulation for a long time,known as AM radio. This modulation scheme uses a fixed carrier frequency,and the amplitude of the signal indicates the information transmitted. For binarytransmission, the ”full” amplitude may represent a 1, and half of that amplitudemay represent a 0. This is illustrated in Figure 3, where the carrier has beenmodulated with the signal in Figure 2.

0 50 100 150 200 250 300 350−1

−0.5

0

0.5

1

Figure 1: An example carrier signal

0 50 100 150 200 250 300 350

0

0.2

0.4

0.6

0.8

1

Figure 2: A data signal


0 50 100 150 200 250 300 350−1

−0.5

0

0.5

1

Figure 3: Amplitude modulation

When phase modulation is used, the phase of the carrier signal is shifted.With our binary example, a 1 may be represented with a phase shift of zerodegrees, while a 0 may be represented by a 180 degree phase shift, as seen inFigure 4.

0 50 100 150 200 250 300 350−1

−0.5

0

0.5

1

Figure 4: Phase shift modulation

The third way of modulation is frequency modulation (which is widely usedin FM-radio broadcasting). Here, the frequency of the carrier signal is varied ac-cording to the data that is to be transmitted. For example, a 1may be representedby a frequency of 800 Hz, while a 0 may be represented by a frequency of 400Hz, as illustrated by Figure 5.

These examples assume transmission of only 0’s and 1’s, by using two dis-tinct modulation levels. By adding more levels, for example 25%, 50%, 75%and 100% of the maximum amplitude, more bits can be represented per timeunit (i.e. per transmitted symbol). This can be increased even further, from 4to 8, 16, or 32 levels and so on. The limit for the number of modulation levels(or modulation order) is dependent on how clear the signal is compared to thenoise1, named the signal to noise ratio. For example, 25% (of the maximumamplitude) may represent 00 and 50% may represent 01. The sender transmits

1The noise may also include interference from the sender itself, or other sources.

2. Data Transmission Background 7

0 50 100 150 200 250 300 350−1

−0.5

0

0.5

1

Figure 5: Frequency modulation

at 25%, but on the path to the receiver the signal may have become amplified to38%. This symbol may then be erroneously demodulated as 01 by the receiver.The better the signal, the more levels can be used, and thus a higher channelcapacity is obtained.

To further improve capacity, different modulation techniques can be com-bined. For example, amplitude and phase modulation is commonly combined.By using two modulation levels for both amplitude and phase, four differentcombinations (symbols) are created, while still being as robust to noise as eachof the modulations themselves. The combination of amplitude and phase modu-lation is commonly referred to as QAM (Quadrature Amplitude Modulation).

The cause for amplification and attenuation is due to the properties of radiowaves. Considering an omni-directional or semi-directional antenna, the radiowaves that reach the receiver may arrive directly from the transmitter. The com-munications is said to be in line of sight. Often there are additional componentsin the received signal, caused by waves that have bounced off buildings, off theground, or off other objects in the path. These extra rays then contribute to afuzziness in the reception, as parts of the signals are delayed and all these sig-nals are combined the at point of reception. In urban terrain, line of sight maybe rare because of buildings and other objects in the path from the transmitterto the receiver. This complicates the detection further, as all received signals arereflections from other objects. This is characteristic of radio transmission thatis non-directional. For example directional laser or microwave point-to-pointwireless transmission, does not experience such problems.

Comparing wireless to wired transmission, the wire is a guided and oftenshielded medium. This means that most of the transmitted energy reaches thereceiver with a low amount of interference. The shield also works the other way- transmissions in a wire does not normally2 cause interference. This means a

2Of course, there is always a magnetic field around a wire with flowing current. Also, the wire


much greater freedom in choosing frequencies to use, compared to the heav-ily regulated radio spectrum for wireless transmission. This then allows for amuch higher bandwidth, with a lower amount of errors, in wired transmissioncompared to wireless transmission.

3 Overview of Research Area

As discussed in the previous sections, moving from wired to mobile wirelesscommunication provides additional problems to handle. In general, these canbe attributed to higher bit error rates, disconnect/handover events, and limitedand varying bandwidth. The proposed optimizations for these problems can becoarsely classified into four groups.

Link layer optimizations adhere to the philosophy that the network shouldprovide the best possible service to the transport layer, by hiding the heterogene-ity. For example “snoop” [2] performs caching and local retransmissions overthe wireless link to avoid triggering congestion avoidance at the sender.

Split connection approaches instead use two connections. The mobile hasone connection to the base station over the wireless link. The base station thenhas a connection to the server, over a wired network. TCP can then be usedbetween the server and base station, and a more suitable protocol can be usedover the wireless link, as is for instance done in WAP [18, 29].

Explicit notification type of optimizations reason that the communicationis best served by having information available at the end-point(s). This way,the sender can for example be able to distinguish congestion from data loss inthe wireless network. In [10], a scheme is proposed where “ICMP-DEFER”messages can be sent by a base station to the sender. The sender can then avoidshrinking the congestion window.

End-to-end optimizations are implemented in one or both of the communi-cating end-points. A common argument is that the end-point is the only placethat can completely and correctly implement a functionality in question [23]. Anexample of this is TCP Westwood [5], that modifies the bandwidth estimation al-gorithm in the sender, which has been shown to improve the throughput in bothwired as well as wireless links.

Another area related to our research is the concept of partial reliability. TheTCP protocol provides full reliability, while the UDP [19] protocol provides noreliability. The partially reliable transport protocol, PRTP [4], is an example of

may act as an antenna. For our discussion these issues are not considered.

4. Research Questions 9

a TCP modification that allows the application to choose its required reliabilitylevel. PRTP operates on a per-packet level. A similar idea, but operating onthe bit-level, is used by TCP-L. Here, the needed reliability (or rather integrity)of received packets can be set by the application. This is inspired by the UDP-Lite [14] protocol, which uses a resembling concept but for UDP.

The research area also expands into experimental network emulation. Emu-lation and simulation has for a long time been important tools to evaluate com-puter networks [1, 8, 9, 16, 17]. Our research continues this tradition, with theemulation of a 4G downlink scenario and emulation of controlled packet lossand bit errors.

To summarize, our research area is focused on experimental evaluations oftransport protocols, investigating link-layer and end-to-end optimizations forwireless networking.

4 Research Questions

The overall research question for this thesis is

• How can performance of the wireless Internet be improved from the trans-port layer point of view?

This is a quite broad question, and it is therefore further divided to put it intocontext. A natural first step is to investigate the current state of the art, leadingto the question

• To which existing wireless networks do problems and proposed optimiza-tions of TCP over wireless networks apply?

We find that there are many suggestions from the research community, butnone that is accepted as a good general solution. All proposed optimizations alsoconserve the semantics of full reliability used in TCP.

Inspired by the work in e.g. UDP-Lite [13] and PRTP [4], a question israised about the reliability in relation to the performance of TCP. This questionis formulated as

• What performance can be expected from a more flexible transport servicewhere reliability may be traded for performance?


For this investigation, we also need to look into the interactions between thetransport layer and the lower layers. The link layer for example can provide afully reliable service at the expense of increased delay, drop erroneous packets,or forward them although erroneous. This raises the question of

• How do the lower layers and the transport layer interact?

5 Research Methodology

To answer the questions in the previous section it is good to follow a methodol-ogy. There are many methods of performing research. Problems can for exam-ple be analyzed analytically, where a system is modeled with mathematics andsolved with equations. Another method is experimental research, where a modelis subjected to specific inputs, and the outputs are studied. This is especiallyuseful in complex environments where an analytical model is too abstract.

The research methodology used for evaluations in this thesis can be catego-rized as experimental network emulation. The conceived ideas have been im-plemented in network test beds. Experiments have then been performed in thesetest beds. Measured output (the metrics) from these experiments can then becompared with the output from experiments with a base-line (standard) config-uration. From such comparisons conclusions can be drawn as to what effect acertain modification or implementation has. It is often impossible, or impracticalat best, to test all combinations of parameter settings and metrics, as these growexponentially for each new parameter. The scope of the evaluated parameters hastherefore been limited by necessity, but is believed to capture relevant aspects ofthe investigated protocols and systems.

6 Main Contributions

The contributions of this thesis consist mainly of three parts.First, a number of TCP optimizations have been summarized, together with a

problem description of TCP in wireless networks. The optimizations have beenrelated to a number of relevant wireless data transmission technologies. Thiscontribution is presented in Paper I.

Second, we have developed a TCP modification that allows the application tochoose a more flexible transport service. The application can, by accepting errorsin its data stream, achieve a higher performance (compared to using regular TCP)in cases where residual bit errors are the cause of packet loss in the receiver. The

7. Paper Summary 11

actual performance improvement is dependent upon the amount of errors in thenetwork. For example, Paper II presents a scenario where TCP-L achieves thedouble throughput of TCP at a residual BER of 10

−5. The intended use case forTCP-L is in a wireless receiver that forwards damaged IP packets, running anapplication that is capable of handling errors in the data stream. The modificationonly needs to be implemented in the receiver, and does not need support fromthe network or the sender. It can be argued whether such a modification canbe compared to other TCP optimizations, or even TCP itself, as the protocolsdiffer in the semantics on transfer reliability. Nevertheless, it allows a TCP-Lreceiver to communicate with a TCP sender. Further, a study of header recoveryin TCP-L is presented in Paper III.

Third, the thesis investigates the interactions between the lower layers andthe transport layer. A network emulator, WIPEMU, has been implemented andconfigured according to the parameters of a 4G system proposal [26]. Exper-iments with TCP, TCP-L and TCP Westwood+ [15] have been performed andare reported in Paper IV. The results indicate that for best effort traffic, a highlypersistent link layer with adaptive modulation should be used to obtain the max-imum throughput. On a similar subject, the thesis further contributes knowledgeon the impact of loss generation on emulation-based protocol evaluation. Bybeing able to precisely place packet loss and bit errors, protocol characteristicscan be studied that are otherwise lost in the averaging process over randomlydistributed losses and errors. This is presented in Paper V.

7 Paper Summary

7.1 Paper I - TCP over Wireless Networks

The first paper provides an overview and background of common wireless tech-nologies used today, with emphasis on data transmission. It further outlinesthe TCP protocol, with a special focus on the mechanisms that are problematicin wireless networks. A number of TCP optimizations and modifications havebeen proposed in the research community. The paper summarizes many of theseproposals in relation to the investigated wireless technologies.

7.2 Paper II - Bit Error Transparent Multimedia Transport

The second paper provides insight into the requirements of multimedia applica-tions, and how they may have different requirements both on network transportreliability and data integrity than traditional applications such as file transfer and


E-mail. The paper introduces the idea of a TCP modification, called TCP-L,that allows bit errors in the data stream. The paper also presents experimentscomparing the throughput of TCP-L to TCP with various amounts of residual biterrors.

7.3 Paper III - TCP-L: Allowing Bit Errors in Wireless TCP

The third paper further expands on the ideas in the previous paper. The maincontribution is a concept for header recovery in TCP-L. A damaged TCP packetmay have errors in the payload, in the header, or in both. Errors in the payloadpart need to be handled by the application, while errors in the header need tobe handled by the transport layer. The paper analyzes the header in detail anddiscusses how it can be recovered to a safe state. The paper also presents someexperimental results on the performance of TCP-L for the case where headerrecovery is always successful.

7.4 Paper IV - Emulation and Validation of a 4G System Proposal

The fourth paper presents an emulator for a proposed 4G downlink specificationfrom the “Wireless IP” project [24]. The emulator is used to examine the im-pact of modulation and retransmissions in the link layer on the transport layer.This evaluation is done in the context of three TCP variants: TCP, TCP-L andTCP-Westwood+. The paper further supports the performance improvements ofTCP-L in presence of residual bit errors. It concludes that the link layer in ques-tion should use highly persistent link layer retransmissions in combination withadaptive modulation.

7.5 Paper V - The Impact of Loss Generation on Emulation-basedProtocol Evaluation

A major part of the thesis work involves experimental network studies. Thefifth paper presents background work on network emulation. The key point ofthe paper is that by being able to control packet loss and bit errors in detail,aspects of network protocols can be studied that are otherwise lost in averaging,assuming a random loss process. The implementation of bit error generationdescribed in this paper has also been used for the experiments reported in PapersII and III.

REFERENCES 13

8 Future Work

There exists a number of continuations of this work. The WIPEMU emulator canbe extended to support multiple users together with channel scheduling, framesoft-combining and forward error coding techniques. This allows for investiga-tion of different resource scheduling algorithms and criteria, such as [7]. It mayalso highlight effects not directly visible in the currently investigated one-userscenario, such as how the total system capacity is affected by different numberof link layer retransmissions and modulation switching levels. Also, until nowonly bulk data transfer has been studied. Short-lived flows, or long-lived flowswith sporadic transmissions, would be good to investigate as these are also com-mon traffic patterns.

The current research focus has been on TCP. Instead of modifying and op-timizing this protocol, new protocols such as DCCP [21] and SCTP [25] areemerging. Are applications perhaps better served by moving on from TCP andinstead use new transport protocols?

With an increased interaction between layers, the concept of cross-layer in-formation is an interesting research area. For example, the transport layer couldhave access to information normally only available to the link or physical layer.“Soft information” from the signal decoding in the physical layer could be usefulwhen the transport layer or application layer is decoding data whose correctnessis uncertain. How should this information be conveyed in a good way?

References

[1] S. Bajaj, L. Breslau, D. Estrin, K. Fall, S. Floyd, P. Haldar, M. Handley,A. Helmy, J. Heidemann, P. Huang, S. Kumar, S. McCanne, R. Rejaie,P. Sharma, K. Varadhan, Y. Xu, H. Yu, and D. Zappala. Improving simula-tion for network research. Technical Report 99-702, University of SouthernCalifornia, March 1999.

[2] H. Balakrishnan, S. Seshan, E. Amir, and R. H. Katz. Improving TCP/IPperformance over wireless networks. In proc. 1st ACM Int’l Conf. on Mo-bile Computing and Networking (Mobicom), November 1995.

[3] C. Bettstetter, H.-J. Vogel, and J. Eberspacher. GSM phase 2+ generalpacket radio service GPRS: Architecture, protocols, and air interface. IEEECommunication Surveys, 2(3), 1999.


[4] A. Brunstrom. Research project proposal: Analysis and implementation ofa partially reliable transport protocol for multimedia applications. ProjectProposal, University of Karlstad, February 1998.

[5] C. Casetti, M. Gerla, S. Mascolo, M. Y. Sanadidi, and R. Wang. TCP West-wood: Bandwidth estimation for enhanced transport over wireless links. InProceedings of ACM Mobicom, pages 287–297, Rome, Italy, July 2001.

[6] H. Elaarag. Improving TCP performance over mobile networks. ACMComputing Surveys, 34(3):357–374, August 2002.

[7] N. C. Ericsson. Revenue Maximization in Resource Allocation. PhD thesis,Signals and Systems, Uppsala University, 2004.

[8] K. Fall. Network emulation in the VINT/NS simulator. Proceedings of thefourth IEEE Symposium on Computers and Communications, July 1999.

[9] K. Fall and S. Floyd. Simulation-based comparisons of Tahoe, Reno andSACK TCP. Computer Communication Review, 26(3):5–21, July 1996.

[10] S. Goel and D. Sanghi. Improving TCP performance over wireless links.Proceedings of TENCON 98 - IEEE Region Ten Conference on Global Con-nectivity in Energy, Computer Communication and Control, 1998.

[11] H. Granbohm and J. Wiklund. GPRS - general packet radio service. Eric-sson Review, (2):82–88, 1999.

[12] V. Jacobson, R. Braden, and D. Borman. RFC 1323: TCP Extensions forHigh Performance, May 1992.

[13] L.-A. Larzon, M. Degermark, and S. Pink. UDP Lite for real-time multi-media applications. Proceedings of the IEEE International Conference ofCommunications (ICC) - Quality of Service Mini Conference, June 1999.

[14] L.-A. Larzon, M. Degermark, S. Pink, L.-E. Jonsson, and G. Fairhurst.RFC 3828: The lightweight user datagram protocol (udp-lite), July 2004.

[15] S. Mascolo, L. A. Grieco, R. Ferorelli, P. Camarda, and G. Piscitelli. Per-formance evaluation of Westwood+ TCP congestion control. PerformanceEvaluation, 55(1-2):93–111, January 2004.

[16] B. Melander and M. Bjorkman. Trace-driven network path emulation.Technical report 2002-037, Department of Information Technology, Up-psala University, Sweden, 2002.

REFERENCES 15

[17] B. Noble, M. Satyanarayanan, G. Nguyen, and R. Katz. Trace-based mo-bile network emulation. In Proceedings of ACM SIGCOMM ’97, Septem-ber 1997.

[18] S. Pehrson. WAP - the catalyst of the mobile Internet. Ericsson Review,(1):14 – 19, 2000.

[19] J. Postel. RFC 768: User Datagram Protocol, Aug. 1980.

[20] J. Postel. RFC 793: Transmission control protocol, Sept. 1981.

[21] E. K. Sally Floyd, Mark Handley. Problem statement for DCCP. draft-ietf-dccp-problem-00.txt, Work in progress, October 2002.

[22] J. H. Saltzer, D. P. Reed, and D. D. Clark. End-to-end arguments in sys-tem design. ACM Transactions on Computer Systems, 2(4):277–288, Nov.1984.

[23] J. H. Saltzer, D. P. Reed, and D. D. Clark. End-to-end arguments in sys-tem design. ACM Transactions on Computer Systems, 2(4):277–288, Nov.1984.

[24] M. Sternad. The wireless IP project, June 2002.

[25] R. Stewart. RFC 2960: Stream Control Transmission Protocol, October2000.

[26] A. Svensson, A. Ahlen, A. Brunstrom, T. Ottosson, and M. Sternad. AnOFDM based system proposal for 4G downlinks. In Proceedings of Multi-Carrier Spread Spectrum Workshop, Oberpfaffenhofen, Germany, 2003.

[27] M. Taferner and E. Bonek. Wireless Internet Access over GSM and UMTS.Springer-Verlag, 2002.

[28] The Working Group for WLAN Standards. IEEE 802.11 wireless localarea networks (webpage). http://grouper.ieee.org/groups/802/11/, checked2005-04-12.

[29] WAP Forum Ltd. Wireless Application Protocol - architecture specifica-tion. http://www.wapforum.org, April 1998.

Division for Information TechnologyDepartment of Computer Science

Karlstad University Studies2005:13

Stefan Alfredsson


and Evaluations

TCP in Wireless Networks: Challenges,Optimizations and Evaluations

This thesis presents research on transport layer behavior in wireless networks. As the Internet is expanding its reach to include mobile devices, it has become apparent that some of the original design assumptions for the dominant transport protocol, TCP, are approaching their limits. A key feature of TCP is the congestion control algorithm, constructed with the assumption that packet loss is normally very low, and that packet loss therefore is a sign of network congestion. This holds true for wired networks, but for mobile wireless networks non-congestion related packet loss may appear. The vary-ing signal power inherent with mobility and handover between base-stations are two example causes of such packet loss. This thesis provides an overview of the challenges for TCP in wireless networks together with a compilation of a number of suggested TCP optimizations for these environments. A TCP modification called TCP-L is proposed. It allows an application to increase its performance, in environments where residual bit errors normally give a degraded throughput, by making a reliability tradeoff. The performance of TCP-L is experimentally evaluated with an implementation in the Linux kernel. The transport layer performance in a 4G scenario is also experimentally investigated, focusing on the impact of the link layer design and its parameterization. Further, for emulation-based protocol evaluations, controlled packet loss and bit error generation is shown to be an important aspect.

Karlstad University StudiesISSN 1403-8099

ISBN 91-85335-53-3

Stefan Alfredsson T

CP in W

ireless Netw

orks: Challenges, O

ptimizations and E

valuations

tcp in wireless networks: challenges, optimizations and ...24680/fulltext02.pdf · during the last...

Documents